Climate science

The 'pause' unpacked

Short-term climate trends are sensitive to definitions, data and testing. This sensitivity underlies an alleged pause in global warming, and highlights the need for meaningful definitions to sustain claims that it was real. See Analysis p.41

The climate system is warming inexorably, but unevenly, in response to increasing concentrations of greenhouse gases in the atmosphere. Although fluctuations in the rate of warming are expected, there has been much research into the characteristics and causes of the most recent period of slower-than-average warming, which occurred from about 1998 for a decade or so. The results of this research have sometimes seemed inconsistent. On page 41, Medhaug et al.1 articulate the choices made by different analysts in studying short-term climate trends, and explain their implications.

Conventionally, research on climate change has not focused on short-term trends (a decade or so in length) because such periods are dominated by natural climate variations, rather than by slower changes in greenhouse-gas concentrations, and therefore don't address larger climate-change issues. However, with some claiming that climate change had somehow 'paused' or entered a 'hiatus', part of the research focus shifted to these short-term trends.

This has posed challenges for climatologists because conventional climate metrics (such as the global-mean surface temperature; GMST) and tools (such as climate-model projections) are not well-attuned to such short time frames. Trends measure rates of change and are sensitive to any uncertainty in the data, as well as to the assumptions used to model the trend. This sensitivity can be demonstrated by comparing the GMST data and GMST trends. For example, the time series of the GMST is nearly identical for the HadCRUT3, HadCRUT4 and GISTEMP data sets2,3, which display similar variation and long-term trends (Fig. 1a). But when the raw data are modelled as short-term trends (measured from 1998), small differences in the data can yield notably different results4 (Fig. 1b).

Figure 1: Climate variability.

a, The time series of the global-mean surface temperature (GMST) is nearly identical for the HadCRUT3, HadCRUT4 and GISTEMP data sets2,3. Here, the data are plotted as a departure (anomaly) from the average value of the GMST during the period 1981–2010. b, But when these data are modelled as trends (measured in monthly increments from 1998), the type of model used and small differences in the data give remarkably different results4. The year labels span the calendar year, with the tick marks corresponding to 1 January. Medhaug et al.1 reconcile these results and show that the climate-model projections and observed trends are consistent. c, GMST warming fluctuates about a longer-term average warming rate — the warming is sometimes faster than the average rate (red dashes) and at other times slower (blue dashes). Some interpretations of the periods of slower-than-average warming include a 'pause' in which the short-term trend is either not positive (no trend) or below the average long-term trend (a slow trend).

Remarkably, the HadCRUT3 data set shows near-zero trends for parts of the slower warming period, whereas the GISTEMP data set displays positive trends of about 0.1 °C per decade or more. Furthermore, the updated version of the HadCRUT3 data set (HadCRUT4) shows trends that are closer to those of GISTEMP than to those of HadCRUT3. The HadCRUT4 data set contains changes to the way sea surface temperature is processed to account for changes in measurement systems, and has better global coverage than HadCRUT3. These differences seem insignificant in terms of the time series, but they substantially change the trends.

Medhaug and colleagues describe changes to the GMST data that bring the trends closer to agreement. These changes include accounting for data-sparse regions such as the Arctic, which was represented in some GMST data sets but not others. Because the rate of Arctic warming seems to have been high (compared with the global average) in the past few decades, the determination of these sensitive short-term trends depends on whether or not the Arctic is included. The role of Arctic warming in GMST trends was highlighted as early as 2008 (see

Another key factor highlighted by Medhaug et al. in unpacking the different claims about whether the GMST paused is how one defines a 'pause' (Fig. 1c). The GMST warming rate fluctuates naturally about a longer-term mean rate, mostly as a result of circulation processes in the ocean that occur on decadal timescales. Early studies5 on this issue typically defined a pause as a period in which there is no significant positive trend in the GMST. However, this is not a useful definition because the data continue to show5 significant warming trends when the trend length exceeds 16 years. Another interpretation is that a pause corresponds to a substantial reduction, or 'slowdown', in the magnitude of the trend. Evidence for such a feature in the data is similarly weak and is not supported by change-point analyses6.

A further definition used for a pause is a departure from climate-model expectations of warming rates. The difficulty here is that projections of the climate's response to greenhouse gases were not designed to assess short-term trends. Such trends are dominated by natural internal climate variations, and the timing of these variations is not synchronized between the model projections and the real world7. Experience in comparing such projections with the observed GMST has shown the need to account for internal variations, and to consider uncertainties in the scenarios for the evolution of emissions and volcanic eruptions supplied to the models. Furthermore, the use of 'blended' sea-surface and air temperatures in the observed GMST, but only air temperatures for the model projections, has led to 'apples and oranges' comparisons that explain part of the difference between the models and the data8. Medhaug et al. show that, when all of these factors are properly accounted for, climate-model expectations of warming rates and observed trends are not at odds.

In short, some data, tools and methods that were good enough when looking at longer-term climate change proved to be problematic when they were focused on the problem of explaining short-term trends. Small differences in GMST data that are inconsequential for climate change are amplified when short-term trends are calculated. Climate-model projections are blunt tools for the analysis of short-term trends. As concluded by Medhaug and colleagues, the different choices and definitions made by analysts underpin the diverging positions on the existence of the pause.

Perhaps the most salient lesson to be learnt from work on the pause is the need for clarity of definition and for quantifiable, generalizable accounts of the alleged phenomenon. Across the hundreds of papers written on the pause, it is hard to find clear definitions that can be usefully generalized. Too often, the pause period defined is so short that a generalization would imply that warming is paused for about one-third of the time5.

The Intergovernmental Panel on Climate Change (IPCC)9 defines the pause as the reduction in GMST trend during 1998–2012 in comparison to 1951–2012, which offers no basis for generalization. For instance, it does not tell us how to quantify such a reduction. If the IPCC definition is to be generalized beyond the nominated 15 years (1998–2012), then we would need to test for all 15-year periods that have a trend below the long-term trend. Because the GMST warming rate fluctuates, the 15-year warming rate will, on average, be paused for about half the time, which is not a useful generalization. Our definitions, like our tools, need sharpening if they are to sustain claims about unusual climate events. Footnote 1


  1. 1.

    See all news & views


  1. 1

    Medhaug, I., Stolpe, M. B., Fischer, E. M. & Knutti, R. Nature 545, 41–47 (2017).

    CAS  ADS  Article  Google Scholar 

  2. 2

    Morice, C. P., Kennedy, J. J., Rayner, N. A. & Jones P. D. J. Geophys. Rev. 117, D08101 (2012).

    ADS  Article  Google Scholar 

  3. 3

    Hansen, J., Ruedy, R., Sato, M. & Lo, K. Rev. Geophys. 48, RG4004 (2010).

    ADS  Article  Google Scholar 

  4. 4

    Rahmstorf, S., Foster, G. & Cahill, N. Environ. Res. Lett. 12, 054001 (2017).

    ADS  Article  Google Scholar 

  5. 5

    Lewandowsky, S., Risbey, J. S. & Oreskes, N. Sci. Rep. 5, 16784 (2015).

    CAS  ADS  Article  Google Scholar 

  6. 6

    Cahill, N., Rahmstorf, S. & Parnell, A. C. Environ. Res. Lett. 10, 084002 (2015).

    ADS  Article  Google Scholar 

  7. 7

    Risbey, J. S. et al. Nature Clim. Change 4, 835–840 (2014).

    ADS  Article  Google Scholar 

  8. 8

    Mann, M. E., Rahmstorf, S., Steinman, B. A., Tingley, M. & Miller, S. K. Sci. Rep. 6, 19831 (2016).

    CAS  ADS  Article  Google Scholar 

  9. 9

    Stocker, T. F. et al. in Climate Change 2013: The Physical Science Basis. Working Group I Contribution to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (eds Stocker, T. F. et al.) 33–115 (Cambridge Univ. Press, 2013).

Download references

Author information



Corresponding authors

Correspondence to James S. Risbey or Stephan Lewandowsky.

Related links

Related links

Related links in Nature Research

Earth science: Free and forced climate variations

Climate science: The cause of the pause

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Risbey, J., Lewandowsky, S. The 'pause' unpacked. Nature 545, 37–39 (2017).

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing